Attention and habit
What a digital detox does
Randomised trials of quitting social media find real but small well-being gains that do not persist, and the strictest meta-analysis of abstinence finds no effect at all.
Taking a break from social media produces a measurable improvement in how people say they feel, and the improvement is small: across the randomised evidence it sits at roughly a fifth of a standard deviation, real but hard to notice in one life. It does not appear to last: where trials followed people afterwards, use returned to baseline within weeks and the mood gain was not shown to survive. The literature also contradicts itself — a meta-analysis restricted strictly to abstinence found no significant effect on positive affect, negative affect or life satisfaction. Almost every trial is unblinded: people know they are detoxing while being asked how they feel. For a workplace-oriented comparison point on how digital activity is translated into metrics, see this guide.
The best-designed experiment found a small gain
The benchmark study paid heavy Facebook users to deactivate for a month.
Deactivation improved subjective well-being by 0.09 standard deviations, freed about 60 minutes a day and cut political polarisation by 0.16 SD; the median willingness-to-accept was $100.
Direction: Increase. Strength of evidence: Strong.
Caveat The sample was young, well-educated, left-leaning and limited to users of 15 minutes a day or more, and the gain is roughly a quarter to two-fifths of the effect of an intervention such as therapy.
It is also the best source on what happens afterwards. One month later the treatment group still used Facebook about 12 minutes a day less, a 23% reduction, and nine weeks on 5% remained deactivated against 2.5% of controls (Allcott et al., American Economic Review, 2020). Those are behavioural measures; persistence of the mood gain was not established.
The meta-analyses disagree with each other
Two syntheses find a small positive effect on well-being. A third, applying a stricter definition of abstinence, finds nothing. The disagreement is genuine, not weak studies against strong ones.
Social media restriction improved well-being by g = 0.17 (95% CI 0.08–0.27), statistically significant but small.
Direction: Increase. Strength of evidence: Strong.
Caveat The authors conclude that future work should target mechanisms rather than broadly restricting time on social media, and most trials were short, unblinded and open to expectancy effects.
Digital detox improved subjective well-being by SMD 0.21 and psychological well-being by 0.27.
Direction: Increase. Strength of evidence: Mixed.
Caveat Heterogeneity was high (I² = 73.6% for subjective well-being), trim-and-fill correction for publication bias was required, and four studies contributed the psychological well-being estimate.
Restricted to abstinence, effects were non-significant throughout: positive affect g = 0.03, negative affect g = −0.01, life satisfaction g = 0.03, with no moderation by duration.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat Only ten studies met the stricter abstinence criterion, and the divergence from the other meta-analyses turns largely on which outcomes and studies are admitted.
A shift of 0.17 to 0.21 standard deviations is a group average, not anyone's month, and it is the same order of magnitude as the 0.09 SD in the deactivation trial — reassuring for the estimate, unhelpful for anyone hoping for a transformation.
Scale it up and the political effects vanish
A larger successor experiment deactivated tens of thousands of accounts before the 2020 US election.
Effects on polarisation, perceived election legitimacy, candidate favourability and voter turnout were precisely estimated and close to zero.
Direction: No detectable effect. Strength of evidence: Strong.
Caveat Run with Meta under a pre-registered independence agreement, but Meta provided the platform data and access — a structural conflict — and it measured political rather than well-being outcomes.
Effects fade as soon as the trial stops
Two 2025 trials measured what happened afterwards and checked compliance rather than trusting it.
Restriction improved depressive symptoms (η² = .109), well-being, stress and sleep quality, but screen time rapidly returned and only insomnia symptoms showed a sustained effect at six-week follow-up.
Direction: Increase. Strength of evidence: Mixed.
Caveat Small, non-blinded, in healthy students with low baseline symptoms, which risks floor effects, and all mental-health outcomes were self-reported.
Blocking mobile internet improved sustained attention (dz = 0.23), mental health (0.56) and subjective well-being (0.45), but only 25.5% of participants met the preregistered compliance threshold.
Direction: Increase. Strength of evidence: Mixed.
Caveat Participants self-selected because they were already worried about their phone use, the intervention could not be blinded, and attrition was higher among those with worse baseline mental health.
A one-week break from Facebook, Instagram, Twitter and TikTok in 154 adults improved well-being, depression and anxiety scores (Lambert et al., Cyberpsychology, Behavior, and Social Networking, 2022), but compliance was self-reported with no verification. It illustrates the design problem running through the literature: the people reporting the benefit are the people who know they are meant to be feeling one.
What the trials support and what they do not
The defensible reading is narrow. Cutting social media for a few weeks shifts average self-reported well-being up slightly, frees an hour or so a day and, in the one trial that measured it well, leaves a modest reduction in use a month later. Sleep is the most durable signal, fitting what controlled studies show about devices and sleep. None of it establishes that heavy use was causing harm in the first place — see whether technology addiction is a real diagnosis — and none supports the claims of retreat operators and app vendors.
The short version
- Randomised deactivation of Facebook for four weeks improved subjective well-being by 0.09 standard deviations and freed about an hour a day (Allcott et al., 2020).
- Two meta-analyses put the well-being gain from restriction at g = 0.17 and SMD 0.21, both small and drawn mostly from short unblinded trials.
- A meta-analysis restricted to strict abstinence found no significant effect on positive affect, negative affect or life satisfaction, so the field disagrees with itself.
- A six-week deactivation of nearly 20,000 Facebook accounts before the 2020 US election found political effects close to zero (Allcott et al., PNAS, 2024), in a study run with Meta's cooperation.
- The evidence cannot say whether benefits outlast the break: the trial that followed participants found use returned to baseline, only insomnia symptoms still improved.